Users who interact with automatic speech recognition systems for interactive media systems often desire to have a natural conversation with the system. For example, a user may issue multiple search queries, and expect the interactive media system to remember previous queries and distinguish between corrections to previous queries and new queries. However, conventional systems lack this natural conversation aspect, being based on a syntactic interpretation of queries which often inaccurately assigns user input as referring to an earlier query, or being a new query. Accordingly, such conventional systems are inconvenient to use, especially when a user hesitates or uses short hand to modify a previous query. Conventional systems are unable to correctly interpret ambiguities in a sentence such as negation elements or expressions which may yield multiple meanings to the sentence. Such conventional systems lack an effective mechanism for correctly analyzing queries within the context of a conversation.